Medweight Academy Internship Program Executive Summary
Obesity Medicine and Diabetes Institute — Prepared March 2026
What Is This Program?
MedWeight is a deployed, AI-powered patient engagement platform serving patients of the Obesity Medicine and Diabetes Institute. It delivers structured behavioral coaching via SMS-based conversational AI, validated clinical screening across 25+ conditions, a 170+ lesson video academy with pathway-filtered retrieval-augmented generation (RAG), and a multi-layered clinical governance infrastructure.
This proposal defines a structured 16-week pre-medical internship designed to advance MedWeight in the areas where the current evidence base identifies the greatest opportunity: deeper clinical content, richer conversational dynamics, tighter safety protocols, and more meaningful outcome measurement — all grounded in peer-reviewed evidence.
Learning Outcomes for the Intern
This internship is designed to provide exposure and skill development that directly supports a pre-medical trajectory:
Clinical AI Systems
Hands-on experience with how AI is deployed in a clinical setting, including prompt engineering, clinical screening logic, governance, and the regulatory considerations that constrain AI behavior in healthcare
Evidence-Based Medicine
Practical application of literature review, clinical guideline interpretation, and translation of research findings into system design decisions
Digital Health
Understanding of how patient-facing digital tools are designed, deployed, measured, and improved — an increasingly critical competency for physicians entering practice in an AI-augmented healthcare environment
Behavioral Medicine
Deep engagement with MI, CBT, and ACT methodologies as implemented in a scalable digital coaching system
Health Equity
Practical experience evaluating and improving health communication for diverse populations, including health literacy optimization and cultural competency assessment
Clinical Documentation
Experience producing clinical-grade documentation including screening protocols, escalation matrices, and outcome measurement specifications
Interdisciplinary Collaboration
Working at the intersection of technology and medicine, translating clinical knowledge into technical specifications and vice versa
MedWeight by the Numbers
Clinical Conditions Screened
Academy Video Lessons
Condition Categories
Weeks of Structured Work
Parallel Work Streams
Retention in Hybrid AI Models
Current Cardiovascular Risk Reports, 2025 — 21-study review of AI-enabled behavioral coaching in obesity.
Why AI Health Coaching — and Why Now?
A 2025 systematic review in Frontiers in Digital Health (35 studies) confirmed that hybrid AI + human coaching approaches show the most clinical promise, with outcomes including weight loss of 0.8 kg to 13.9% of baseline, systolic blood pressure reductions up to 18.6 mmHg, and HbA1c improvements up to 1.2 percentage points. A 2023 Obesity Medicine Association Clinical Practice Statement established that effective AI chatbots must be personalized, contextualized, immersive, and empathic.
Despite this progress, the literature consistently identifies critical gaps that this internship directly targets: declining user engagement over time, insufficient clinical content depth, limited gamification, algorithmic bias in underserved populations, and the absence of rigorous outcome measurement frameworks.
📈 Declining Engagement
The most persistent challenge across all AI coaching platforms — addressed through gamification, milestone recognition, and motivational arc prompting.
📖 Clinical Content Depth
Conditions detected by screening but underserved by educational content represent a critical RAG pipeline gap requiring evidence-based supplementation.
🛡 Safety Protocols
Beyond suicidal ideation, high-severity binge eating, alcohol use, and medication contra-indications require structured escalation logic.
🌐 Health Equity
Algorithmic bias and health literacy gaps risk leaving underserved populations behind — a pre-med perspective is uniquely positioned to address this.
Seven Integrated Work Streams
The internship is organized into seven parallel work streams, each producing concrete, production-ready deliverables reviewed by the Institute's clinical team before integration.
16-Week Timeline (15–20 hrs/week)
Weeks 1–2 · Onboarding & Orientation
Platform walkthrough, codebase review, literature review of AI health coaching best practices, documentation standards.
Weeks 3–4 · Content Gap Analysis
Map conditions to available content, audit Academy transcripts for clinical accuracy, identify priority gaps.
Weeks 5–6 · Screening Ruleset Enhancement
Expand keyword library, add new conditions, update medication lists, develop detection sensitivity audit.
Weeks 7–8 · Knowledgebase Expansion
Develop supplementary content documents, GLP-1 reference, pathway FAQs, and new questionnaire instruments.
Weeks 9–10 · Prompt Engineering
Develop pathway prompt contexts, condition-aware modifiers, stage-aware layers, and coaching technique cards.
Weeks 11–12 · Safety, Escalation & Cultural Competency
Escalation protocol matrix, contra-indicated advice reference, health literacy audit.
Weeks 13–14 · Outcome Measurement & Gamification
Patient outcome dashboard spec, engagement score algorithm, gamification system design.
Weeks 15–16 · Integration & Final Report
Compile all deliverables, write final summary report, present findings to clinical and technical team.
Success Metrics
Number of conditions in the clinical detection library expanded from 16 to 25+, with documented clinical evidence for each addition
Detection sensitivity audit demonstrating measurable improvement in false-negative rate against synthetic patient messages
5+ new pathway-specific system prompt contexts developed, clinically reviewed, and ready for deployment
10+ supplementary content documents produced and formatted for RAG ingestion
4–6 new validated screening instruments defined and ready for deployment in the questionnaire system
Complete escalation protocol matrix covering all condition categories and severity levels
Gamification system design with 20+ badge definitions and trigger logic
Health literacy audit completed with specific, actionable revision recommendations
All deliverables produced with supporting clinical citations and formatted for immediate integration into the production system
Supervision & Reporting Structure
Weekly Check-In
Weekly 30-minute check-in with the CTO/platform lead to review progress, resolve blockers, and align priorities
Biweekly Deliverable Review
Each two-week cycle produces a defined set of documents that are reviewed for clinical accuracy and technical feasibility before the next cycle begins
Shared Repository
All work is documented in a shared project repository with clear version control
Clinical Review
Clinical content produced by the intern is reviewed by the Institute's clinical team before integration into the production system
Final Presentation
A final presentation to the clinical and technical team at the conclusion of the internship summarizing all contributions, findings, and recommendations for future work
Positioned at the Frontier of Clinical AI
This internship positions a pre-medical student at the frontier of clinical AI — contributing to a production system that serves real patients, governed by real clinical accountability standards, and backed by a growing evidence base that AI health coaching can deliver meaningful clinical outcomes.
MW · MedWeight
MedWeight · Intern Work Proposal
MedWeight Academy Internship Program
Pre-Medical Internship — Clinical AI & Digital Health
Obesity Medicine and Diabetes Institute
Purpose and Context
This proposal defines a structured internship program for a pre-medical student to contribute meaningfully to MedWeight, a production clinical AI platform serving patients of the Obesity Medicine and Diabetes Institute. The work is designed to leverage the intern's clinical knowledge trajectory while providing hands-on exposure to the intersection of artificial intelligence, behavioral medicine, and digital health systems.
The intern's work will span clinical content development, AI prompt engineering, screening ruleset refinement, knowledgebase expansion, and evidence-based feature enhancement — all areas where a pre-medical perspective is uniquely valuable and where the current evidence base on AI health coaching identifies significant opportunities for improvement.
MedWeight Platform Capabilities
- SMS-based conversational AI with structured coaching using MI/CBT/ACT methodology
- Validated clinical screening across 25+ conditions
- Closed-loop questionnaire system with longitudinal progress tracking
- 170+ lesson video academy with pathway-filtered retrieval-augmented generation
- Multi-layered clinical governance infrastructure including cryptographic message integrity verification
Learning Outcomes for the Intern
This internship is designed to provide exposure and skill development that directly supports a pre-medical trajectory:
Clinical AI Systems
Hands-on experience with how AI is deployed in a clinical setting, including prompt engineering, clinical screening logic, governance, and the regulatory considerations that constrain AI behavior in healthcare
Evidence-Based Medicine
Practical application of literature review, clinical guideline interpretation, and translation of research findings into system design decisions
Digital Health
Understanding of how patient-facing digital tools are designed, deployed, measured, and improved — an increasingly critical competency for physicians entering practice in an AI-augmented healthcare environment
Behavioral Medicine
Deep engagement with MI, CBT, and ACT methodologies as implemented in a scalable digital coaching system
Health Equity
Practical experience evaluating and improving health communication for diverse populations, including health literacy optimization and cultural competency assessment
Clinical Documentation
Experience producing clinical-grade documentation including screening protocols, escalation matrices, and outcome measurement specifications
Interdisciplinary Collaboration
Working at the intersection of technology and medicine, translating clinical knowledge into technical specifications and vice versa
AI Health Coaching: Relevance and Opportunity in Today's Healthcare Environment
The research landscape on AI-enabled behavioral coaching in obesity and chronic disease management is rapidly maturing. A 2025 systematic review in Frontiers in Digital Health synthesizing 35 studies found that both human and AI coaching modalities demonstrated feasibility and acceptability, with hybrid approaches (AI + human) showing the most promise but requiring further refinement. A 2025 review in Current Cardiovascular Risk Reports examining 21 studies on AI-enabled behavioral coaching in obesity found clinically significant outcomes: weight loss ranging from 0.8 kg to 13.9% of baseline, systolic blood pressure reductions up to 18.6 mmHg, HbA1c improvements up to 1.2 percentage points, and retention rates of 57–92% in hybrid models. A 2023 Obesity Medicine Association Clinical Practice Statement established that AI chatbots should be designed to be personalized, contextualized, immersive, and empathic to enhance engagement and behavior change. However, the literature consistently identifies critical gaps:
- Declining user engagement over time is the most persistent challenge across all AI coaching platforms
- Limited clinical validation through rigorous randomized controlled trials with adequate sample sizes and follow-up periods
- Insufficient integration of gamification, empathy expression, and personification in conversational design
- Algorithmic bias concerns, particularly in underserved populations
- The need for evidence-based content that goes beyond generic advice to clinically grounded, condition-specific guidance
- Opportunity for AI to bridge the gap between clinic visits through continuous behavioral monitoring and just-in-time interventions
MedWeight's Position in the Evidence Landscape
MedWeight is already ahead of most platforms described in the literature — it has structured coaching, validated screening instruments, longitudinal tracking, and governance infrastructure. The intern's work targets the areas where the evidence says the greatest gains remain: deeper personalization, richer clinical content, more engaging conversational dynamics, and tighter alignment between AI behavior and clinical best practices.
Max baseline weight loss in reviewed studies
mmHg max SBP reduction
Max HbA1c improvement
Retention rates in hybrid models
Clinical Screening — Keyword Library Expansion and Refinement
The platform's clinical detection engine operates through two layers: AI-generated structured JSON screening and a PHP keyword-based ClinicalFlagAnalyzer scanning against 25+ conditions across 10 categories. Both layers present significant opportunities for a pre-med intern with clinical knowledge.
Audit & Expand Condition Library
Audit the existing keyword library in clinical_flags.php against current clinical literature and DSM-5-TR criteria. The current library covers 16 conditions; expand to 30+ by adding conditions commonly comorbid with obesity: hypothyroidism, polycystic ovary syndrome (PCOS), non-alcoholic fatty liver disease (NAFLD), obstructive hypoventilation, metabolic syndrome as a distinct entity, peripheral neuropathy, and lymphedema
Expand Medication Brand-Name Recognition
Expand the medication brand-name recognition lists for each condition. Current lists are strong but incomplete — for example, newer GLP-1 combinations (tirzepatide, retatrutide), newer antidepressants (vortioxetine/Trintellix, vilazodone/Viibryd), and newer sleep medications (lemborexant/Dayvigo, suvorexant/Belsomra) should be added
Add Lay Language Pattern Detection
Add lay language pattern detection for conditions currently dependent on medical terminology. Patients rarely say "insomnia" — they say "I'm up at 3am every night staring at the ceiling." Build expanded colloquial phrase sets for each condition
Document Severity Gradation Criteria
Research and document severity gradation criteria for each condition, mapping patient language patterns to low/medium/high confidence levels with clinical rationale for each classification
Clinical Screening — Condition Map Logic Enhancement
Review Condition-to-Questionnaire Mappings
Review the current condition-to-questionnaire mappings in the active condition map. Evaluate whether additional validated instruments should be mapped (e.g., PSQI for sleep quality, EDE-Q for eating disorders, PCL-5 coverage for PTSD which is already deployed, and the Epworth Sleepiness Scale for daytime somnolence)
Research Condition Exclusion Pairs
Research and propose condition exclusion pairs for the clinical ruleset. Currently the exclusion logic is available but sparsely populated. Identify condition pairs where co-flagging is clinically misleading (e.g., flagging both "chronic fatigue" and "depression" when the fatigue is a depression symptom, not a separate condition)
Develop Symptom Weight Recommendations
Develop symptom weight recommendations for the clinical ruleset, documenting the clinical evidence for why certain symptoms are stronger indicators than others (e.g., anhedonia is a stronger depression signal than "feeling tired")
Clinical Screening — Deliverables
Updated clinical_flags.php
Updated clinical_flags.php with expanded conditions, keyword sets, and medication lists, with a citation document providing clinical evidence for each addition
Condition Map Enhancement Proposal
Condition map enhancement proposal with new instrument mappings, exclusion pairs, and symptom weights, reviewed against current clinical guidelines
Detection Sensitivity Audit Report
A "Detection Sensitivity Audit" report: test the current system against 50+ synthetic patient messages representing common presentations and document false negatives
Prompt Engineering — General Support Prompt Enhancement
The AI's conversational quality is governed by the system prompt architecture in twilio_webhook.py (general support) and coaching_sessions.py (coaching mode). The literature consistently identifies personalization, empathy expression, and sustained engagement as the key differentiators between effective and ineffective AI health coaching. This work stream targets all three.
Pathway-Specific System Prompt Contexts
Research and draft pathway-specific system prompt contexts for conditions not currently covered. The current GLP-1 and General pathways are well-developed; create equally detailed prompt contexts for: Diabetes Management, Mental Health, Bariatric Pre-Surgical, Bariatric Post-Surgical, and Nutrition Focus pathways
Conversational Memory Prompt Enhancement
Develop a "conversational memory" prompt enhancement that instructs Claude to reference the patient's stated goals, concerns, and milestones from previous conversations. The current system injects conversation history but does not instruct the AI to actively recall and reference specific patient statements
Condition-Aware Prompt Modifiers
Draft condition-aware prompt modifiers: when the screening JSON detects a specific condition, the system prompt for subsequent interactions should adapt. A patient flagged for depression should receive an AI that is more actively validating and less prescriptive about exercise; a patient flagged for binge eating should receive an AI that avoids food-restriction language
Stage-Aware Prompt Layer
Create a "stage-aware" prompt layer that adjusts conversational depth and focus based on the patient's journey stage. Week 1 patients need reassurance and basic education; Month 3 patients need deeper behavioral coaching and self-efficacy reinforcement
Prompt Engineering — Coaching Session Prompt Enhancement
Session-Number-Aware Coaching Prompts
Develop session-number-aware coaching prompts. Session 1 should be exploratory and goal-setting; Sessions 2–4 should build on commitments and introduce CBT worksheets; Sessions 5+ should focus on maintenance and relapse prevention. The current coaching prompt is static regardless of session count
Structured Coaching Session Templates
Create structured coaching session templates for each pathway module: opening reflection, skill introduction, guided exercise, commitment setting, and homework assignment. These templates should reference specific Academy content
Coaching Technique Cards
Draft "coaching technique cards" — structured prompt segments that introduce specific therapeutic techniques at appropriate points: values clarification exercises (ACT), thought records (CBT), decisional balance worksheets (MI), mindfulness micro-exercises, and behavioral activation planning
Session Notes Summarization Prompt
Research and propose a session notes summarization prompt that generates structured notes capturing: presenting concerns, techniques used, commitments made, homework assigned, and clinical observations. These notes feed into the next session's opening reflection
Prompt Engineering — Engagement, Retention & Deliverables
Engagement and Retention Prompt Strategies
Motivational Arc Prompts
Design "motivational arc" prompts that shift tone and technique across the patient journey, informed by the transtheoretical model (stages of change): pre-contemplation (empathize, don't push), contemplation (explore ambivalence), preparation (build confidence), action (support with specifics), maintenance (prevent relapse)
Celebration & Milestone Recognition
Draft celebration and milestone recognition prompt language for the content drip campaigns, questionnaire completions, coaching attendance streaks, and journey stage advancement
AI Self-Disclosure Research
Research the role of "self-disclosure" by AI coaches (e.g., "Many of the people I work with feel exactly the same way at this point") and draft appropriate prompt language that builds rapport without misrepresenting AI capabilities
Deliverables
- A Prompt Enhancement Specification document containing all proposed prompt changes with clinical rationale, formatted as exact text ready for integration into the codebase
- Five fully developed pathway-specific system prompt contexts with supporting clinical citations
- A coaching technique card library (10+ cards) with structured prompt segments and usage guidelines
Knowledgebase — Content Gap Analysis
The platform's RAG pipeline currently draws from 170+ Academy video transcripts. This is a strong foundation, but the literature on AI health coaching consistently identifies the depth and clinical accuracy of the knowledgebase as a primary determinant of trust and sustained engagement.
Systematic Content Gap Analysis
Conduct a systematic content gap analysis mapping every condition in the clinical_flags.php library against available Academy content. Identify conditions that are detected but for which no educational content exists in the RAG pipeline (e.g., GERD management, chronic pain coping, alcohol reduction strategies)
Academy Transcript Clinical Accuracy Audit
Audit the Academy transcripts for clinical accuracy against current evidence-based guidelines (ADA Standards of Care 2026, ACC/AHA guidelines, ASMBS bariatric guidelines). Flag outdated recommendations, missing nuance, or areas where the evidence has evolved since the content was produced
Breadcrumb/Navigation Structure Evaluation
Evaluate the breadcrumb/navigation structure for completeness and usability. Do patients understand "Weight Management Mastery > Trailblazing Change (8 weeks) > Session 4: Building Self-Efficacy" as a meaningful location reference?
Knowledgebase — Supplementary Content Development
Educational Content Summaries
Develop written educational content summaries (500–1000 words each) for the top 10 conditions detected by the clinical screening system but underserved by current Academy content. These summaries should be evidence-based, written at a 6th–8th grade reading level, and structured for chunking and RAG retrieval
GLP-1 Medication Quick Reference
Create a GLP-1 Medication Quick Reference knowledge document covering: all currently approved GLP-1 and GLP-1/GIP agonists (semaglutide, tirzepatide, liraglutide, dulaglutide, exenatide, and emerging agents), common side effects by dose tier, dose titration schedules, injection technique best practices, protein intake targets, hydration guidelines, and when to contact the clinic. This document addresses the most common patient questions and fills a critical RAG gap
Pathway FAQ Documents
Draft FAQ documents for each coaching pathway module: the 15–20 most common questions patients ask on that pathway, with clinically accurate answers. These become high-priority RAG retrieval targets
Crisis Resource Reference Document
Research and compile a Crisis Resource Reference document that goes beyond the current 988 Lifeline and Crisis Text Line to include: provincial/state-specific crisis lines, specialized eating disorder hotlines, substance use helplines, and domestic violence resources. Map each resource to the conditions that should trigger its delivery
Knowledgebase — Questionnaire Instruments & Deliverables
Questionnaire Instrument Expansion
Research and propose 4–6 additional validated screening instruments for deployment in the questionnaire system. Candidates: Epworth Sleepiness Scale (ESS), Pittsburgh Sleep Quality Index (PSQI), Eating Disorder Examination Questionnaire (EDE-Q), CAGE-AID (substance use), Oslo Social Support Scale (OSS-3), and the WHO-5 Well-Being Index
For each proposed instrument, produce: the full question set formatted to match standard_questionnaires.json schema, scoring logic, severity thresholds with clinical citations, score_direction, estimated completion time, and the clinical rationale for its addition to the platform
Deliverables
Content Gap Analysis Report
Mapping conditions to available content and identifying deficiencies
10 Supplementary Educational Content Documents
Formatted for RAG ingestion
GLP-1 Quick Reference & Pathway FAQs
GLP-1 Medication Quick Reference and pathway FAQ documents
4–6 New Questionnaire Instrument Definitions
In standard_questionnaires.json format with supporting clinical documentation
Academy Content Accuracy Audit Report
With flagged items and recommended corrections
Patient Outcome Measurement and Analytics Enhancement
The literature is unambiguous: sustained engagement declines over time in AI health coaching platforms, and the best mitigation is visible outcome feedback. MedWeight's questionnaire progress tracking is a strong start, but there are significant opportunities to deepen outcome measurement.
Patient Outcome Dashboard Specification
Design a Patient Outcome Dashboard specification: a single-page view (for admin and optionally patient-facing) that integrates questionnaire score trends, coaching session attendance rate, engagement frequency over time, content completion percentage, and journey stage progression. Define the data sources, calculation logic, and visualization approach for each metric
Patient Engagement Score Algorithm
Research and propose a "Patient Engagement Score" algorithm that combines: message frequency, response latency, coaching session attendance, questionnaire completion rate, content link clicks, and conversation depth (measured by average message length). This composite score would enable clinicians to identify disengaging patients before they drop off
Clinical Outcome Proxies
Define clinical outcome proxies that can be derived from conversational data without requiring lab values: self-reported energy level trends, mood trajectory based on sentiment analysis of patient messages, medication adherence signals (mentions of missed doses, running low), and behavioral goal completion rates
Clinical Summary Report Template
Propose a "Clinical Summary Report" template that can be auto-generated for a patient's chart: a one-page document summarizing AI interactions, screening flags, questionnaire results, coaching attendance, and engagement metrics over a defined period. This addresses the insurance/compliance documentation opportunity identified in the executive summary
Deliverables
- Patient Outcome Dashboard specification with wireframes, data source mapping, and calculation definitions
- Patient Engagement Score algorithm design with clinical rationale and scoring weights
- Clinical Summary Report template design
Safety and Escalation Protocol Enhancement
Patient safety is the non-negotiable foundation. The current suicidal ideation detection pathway is robust, but there are additional safety domains that warrant development.
Escalation Protocol Matrix
Research and design an escalation protocol matrix: for each severity level (critical, high, medium, low) across each condition category, define the expected system behavior. Critical/suicidal ideation triggers immediate crisis resources and require_review flags. What should high-severity binge eating trigger? What about high-severity alcohol use? Document the clinical rationale for each escalation action
Contra-Indicated Advice Reference
Develop a "contra-indicated advice" reference for the AI system prompt: specific clinical scenarios where the AI must NOT provide certain types of guidance. For example, a patient on lithium should not receive advice to increase water intake without caveats about lithium toxicity; a patient post-bariatric surgery should not receive standard protein target advice (their targets are different). Map these contra-indications to detectable patient attributes
Provider Notification Enhancement
Design a provider notification enhancement: currently, clinical flags are visible on the dashboard but do not trigger active notifications. Propose a notification specification: which flag combinations should trigger an email/SMS to the care team, what information should the notification contain, and what is the expected response workflow
Adverse Event Reporting Framework
Research adverse event reporting requirements for AI-driven clinical tools and propose a framework for capturing, documenting, and reviewing any instance where the AI's response may have been clinically inappropriate
Deliverables
- Escalation Protocol Matrix document covering all condition categories and severity levels
- Contra-indicated Advice Reference for system prompt integration
- Provider Notification Specification
- Adverse Event Reporting Framework proposal
Gamification and Behavioral Reinforcement System Design
The literature identifies gamification as one of the most underutilized mechanisms for sustaining engagement in AI health coaching. MedWeight has a gamification feature toggle and a milestone badge system for questionnaire completions, but the full potential is unrealized.
Achievement/Badge System
Design a comprehensive achievement/badge system: propose 20–30 badges across categories such as Engagement (consecutive days active, message milestones), Education (content completion streaks, module completions), Coaching (session attendance streaks, homework completion), Screening (questionnaire completions, improvement milestones), and Wellness (self-reported wins). For each badge, define: name, icon, trigger condition, notification message, and clinical purpose
Streak Mechanism
Research and propose a "streak" mechanism: consecutive days of engagement, consecutive coaching sessions attended, consecutive weeks of questionnaire completion. Define the streak display, break/recovery rules, and motivational messaging
Progress Summary Message Template
Design a progress summary message template for weekly or biweekly automated delivery: "This week you completed 2 coaching sessions, watched 3 academy videos, and your PHQ-9 score improved by 2 points. Amazing work!" Define the data sources, calculation logic, and delivery channel
Peer Comparison Feature
Propose a "peer comparison" feature design (with appropriate privacy controls): anonymized, aggregate messaging like "85% of patients at your stage find that protein shakes help with morning nausea" to provide social proof without exposing individual data
Deliverables
- Gamification System Design Document with 20–30 badge definitions, trigger logic, and clinical rationale
- Weekly Progress Summary message template specification
- Peer comparison feature design with privacy framework
Cultural Competency and Health Literacy Optimization
The literature highlights algorithmic bias and health equity as critical concerns in AI health coaching. A pre-med intern can contribute meaningfully to ensuring MedWeight serves diverse patient populations effectively.
Health Literacy Audit
Conduct a health literacy audit of all patient-facing text: SMS messages, questionnaire instructions, progress page content, coaching session language. Evaluate against the CDC's Clear Communication Index and recommend revisions to ensure content is accessible at a 6th–8th grade reading level
Culturally-Adapted Health Communication
Research culturally-adapted health communication best practices for the Institute's patient demographics. Propose prompt modifications that enable the AI to adjust communication style, food examples, exercise recommendations, and cultural references based on patient context
Screening Keyword Library Bias Evaluation
Evaluate the current screening keyword library for cultural and linguistic bias: are the lay language patterns representative of how diverse patient populations describe their symptoms? Propose expanded keyword sets that reflect linguistic variation
Multilingual Readiness Assessment
Research and propose a multilingual readiness assessment: what would be required to extend MedWeight to French (for Canadian compliance) or Spanish? Identify the content, prompts, and system messages that would need translation and propose a prioritized implementation roadmap
Deliverables
- Health Literacy Audit Report with specific revision recommendations
- Cultural Competency Enhancement Recommendations for prompts and content
- Multilingual Readiness Assessment
Proposed Timeline (16 Weeks)
The following timeline assumes a part-time commitment of 15–20 hours per week. Work streams are sequenced so that foundational research informs later deliverables.
Supervision and Reporting Structure
Weekly Check-In
Weekly 30-minute check-in with the CTO/platform lead to review progress, resolve blockers, and align priorities
Biweekly Deliverable Review
Each two-week cycle produces a defined set of documents that are reviewed for clinical accuracy and technical feasibility before the next cycle begins
Shared Project Repository
All work is documented in a shared project repository with clear version control
Clinical Team Review
Clinical content produced by the intern is reviewed by the Institute's clinical team before integration into the production system
Final Presentation
A final presentation to the clinical and technical team at the conclusion of the internship summarizing all contributions, findings, and recommendations for future work
Success Metrics
Number of conditions in the clinical detection library expanded from 16 to 25+, with documented clinical evidence for each addition
Detection sensitivity audit demonstrating measurable improvement in false-negative rate against synthetic patient messages
5+ new pathway-specific system prompt contexts developed, clinically reviewed, and ready for deployment
10+ supplementary content documents produced and formatted for RAG ingestion
4–6 new validated screening instruments defined and ready for deployment in the questionnaire system
Complete escalation protocol matrix covering all condition categories and severity levels
Gamification system design with 20+ badge definitions and trigger logic
Health literacy audit completed with specific, actionable revision recommendations
All deliverables produced with supporting clinical citations and formatted for immediate integration into the production system
Positioned at the Frontier of Clinical AI
This internship positions a pre-medical student at the frontier of clinical AI — contributing to a production system that serves real patients, governed by real clinical accountability standards, and backed by a growing evidence base that AI health coaching can deliver meaningful clinical outcomes. The work is designed to be immediately valuable to the platform while providing the intern with experience that is increasingly differentiating for medical school applications and for a career in a healthcare system where AI will be a standard clinical tool.
MW · MedWeight